OpenAI and Ironclad train GPT-6 Astra on contracting work

OpenAI announced a research partnership with Ironclad to train and evaluate agents on contracting workflows. GPT-6 Astra is the company's first frontier model trained on these tasks, according to a post published on October 6, 2026.

What the research measured

The collaboration is the first in a program in which OpenAI wants to work with a small number of software companies. The stated goal is to make the model understand business rules, run multi-step workflows, and check that the result meets the original request. Ironclad employees and internal OpenAI users helped define 11 legal, commercial, and procurement tasks, such as setting up nondisclosure agreements, creating procurement approvals, and updating a clause based on the selected jurisdiction.

Each task was scored against 8 to 50 criteria. OpenAI estimates an experienced user would take 30 to 40 minutes per task on average. The models trained in hosted product environments and on synthetic tasks derived from public contracts in the SEC's EDGAR database, after filters meant to remove personal data. The company says it did not use OpenAI customer data, OpenAI's internal contracts, or nonpublic Ironclad customer contracts.

The figures are from an internal evaluation

On the research evaluation, Astra with Max reasoning scored 55.0% on average, against 41.6% for GPT-5.6 Sol with High reasoning. Estimated time per attempt fell from 37.0 to 19.2 minutes. OpenAI describes the gain as 32% higher on the score and 48% less time than Sol. An internal model used while developing Astra reached 63.7%, and the company says it wants to bring that gain to future models.

Those times are simulated estimates, not measured customer savings. The 11 tests also do not cover every Ironclad workflow. CTO Sunita Verma said that, for AI to be useful in complex contracting, an agent needs to understand the full cycle and preserve the controls teams rely on. OpenAI opened a form for other software companies to propose similar collaborations.

Sources

Transparency: This content was created, edited, or reviewed with the help of artificial intelligence. The information was cross-checked with public posts on X and sources available on the internet. Check the original sources for the full context.

By GeekikiBot